Characterizing and Identifying Risk for Falls in the LEAPS Study
A Randomized Clinical Trial of Interventions to Improve Walking Poststroke
Background and Purpose—Better understanding of fall risk poststroke is required for developing screening and prevention programs. This study characterizes falls in the Locomotor Experience Applied Post-Stroke (LEAPS) randomized clinical trial, describes the impact of 2 walking recovery interventions on falls, and examines the value of clinical assessments for predicting falls.
Methods—Community-dwelling ambulatory stroke survivors enrolled in LEAPS were assessed 2 months poststroke. Falls were monitored until 12 months poststroke and participants were characterized as multiple or injurious (M/I); single, noninjurious; or nonfallers. Incidence and time to M/I falls were compared across interventions (home exercise and locomotor training initiated 2 months [early-LTP] or 6 months [late-LTP] poststroke). Predictive value of 2-month clinical assessments for falls outcome was assessed.
Results—Among the 408 participants, 36.0% were M/I, 21.6% were single, noninjurious, and 42.4% were nonfallers. Most falls occurred at home in the first 3 months after assessment. Falls incidence was highest for those with severe walking impairment who received early-LTP (P=0.025). Berg Balance Scale score ≤42/56 was the single best predictor of M/I falls.
Conclusions—As individuals with stroke improve in walking capacity, risk for M/I falls remains high. Individuals walking <0.4 m/s are at higher risk for M/I falls if they receive early-LTP training. Berg Balance Scale score at 2 months poststroke is useful for informing falls risk, but it cannot account for the multifactorial nature of the problem. Falls prevention in stroke will require multifactorial risk assessment and management provided concomitantly with exercise interventions to improve mobility.
Despite recent advances, stroke remains the most common disabling neurological condition among American adults.1 Falls are a common complication after stroke.2,3 Between 40% and 70% of individuals fall within 12 months poststroke,3,4 and their hip fracture risk is doubled.5 Individuals with stroke are more likely to become repeat fallers than are those in the general elderly population,3 with incidence of multiple (>1) falls between 42% and 57% in the first year.3,6,7 Recurrent fallers with stroke have greater deficits in mobility and activities of daily living function than do single fallers.8 Although multiple fallers are not at higher risk for injury for any given fall, cumulative injury risk increases with each fall.9 Thus, falls prediction and management for individuals poststroke should focus on multiple falls.4,7,8
A primary goal of stroke rehabilitation is to improve individuals' mobility in the presence of motor, balance, and visual-spatial deficits. Yet, increasing mobility and physical activity increases exposure to fall risks.3 A systematic review of exercise in older people suggests that strength and balance exercises reduce falls, whereas walking training alone may increase them.10 The authors speculate that this may be associated with either increased risk during walking practice, or that time spent walking reduces time available for balance training. Few studies comparing interventions to improve walking recovery poststroke have reported falls.11–14
Most studies of falls in community-dwelling stroke survivors are relatively small and do not consider concurrent effects of physical therapy programs or increased mobility on falls risk. In contrast, the Locomotor Experience Applied Post-Stroke (LEAPS) Phase 3 multisite randomized clinical trial (RCT), which compared rehabilitation programs to promote walking recovery after disabling stroke, provides a unique opportunity to characterize prospectively and to examine risk for falls in the context of 2 walking rehabilitation programs in a relatively large cohort. The purposes of this study were to characterize the incidence and consequences of falls among participants in the LEAPS RCT, to examine the impact on falls of 2 interventions to improve walking recovery, and to assess the value of clinical assessments at 2 months for predicting falls outcome at 12 months poststroke.
Four hundred eight individuals enrolled in the LEAPS multisite RCT were included in this study.15 Inclusion criteria included stroke in the last 45 days, residual paresis, ability to walk 10 feet with no more than 1-person assistance, ability to follow a 3-step command, and self-selected walking speed slower than 0.8 m/s.15,16 Ethics review boards at all participating centers approved the trial protocol, and all participants provided written informed consent.
Participants were assessed by trained, blinded assessors at 2 months poststroke using standardized protocols reported previously.16 From this assessment, we selected 41 variables deemed potentially relevant to 12-month falls outcome, based on previous studies and clinical judgment, to include in this analysis (full list available online-only, Supplemental Table).
Physical Therapy Interventions
At 2 months poststroke, participants were randomly assigned to 1 of 3 treatment groups: a specialized locomotor training program (LTP) that included stepping on a treadmill with body weight support, followed by walking practice over ground delivered early (early-LTP, 2 months poststroke) or late (late-LTP, 6 months poststroke); or a progressive strength and balance exercise program provided by a physical therapist in the home (HEP) initiated 2 months poststroke.15,16 Each intervention was provided for 30 to 36 sessions over 12 to 16 weeks by trained physical therapists.16 Participants were stratified by moderate (0.4–<0.8 m/s) or severe (<0.4 m/s) walking speed impairment. The late-LTP group received only usual care physical therapy based on current practice between 2 and 6 months poststroke and crossed over to LTP at 6 months.
Falls incidence was monitored between 2 and 12 months poststroke. We used international standards for defining and reporting falls,17 including the following definition for a fall: “A person has a fall if they end up on the ground or floor when they did not expect to. Most often a fall starts while a person is on their feet, but a fall could also start from a chair or bed. If a person ends up on the ground, either on their knees, their belly, their side, their bottom, or their back, they have had a fall.” This explanation was provided to participants and caregivers and was printed on monthly calendars issued at randomization. Participants and/or caregivers placed an “X” on the corresponding date if a fall occurred, and they mailed calendars to their study site each month (even if no falls occurred). Study personnel provided reminders as needed.
Participants were contacted by phone to follow-up on reported falls using a standard questionnaire. Information collected for each fall included presence and nature of any injury, location of the fall, and ability to get up independently after the fall. Three categories were used to characterize falls outcome at 12 months poststroke: multiple or injurious (M/I); single, noninjurious (S/NI); and nonfallers. Injurious falls were those resulting in serious injury: fracture, loss of consciousness, or hospital admission.
Conventional statistics were conducted using SAS 9.2 (SAS Institute, Inc). Chi-square tests and analysis of variance were used to assess univariate associations between clinical assessments at 2 months and faller category; alpha value was 0.05. Chi-square tests were used to examine associations between intervention groups and faller categories. A log-rank test was used to compare the probability of M/I fall onset over time across intervention groups. The Classification and Regression Tree method (CART 6.0, Salford Systems) was used to establish a prediction model for 2-month clinical assessment variables and faller category at 12 months.18 Because we had 3 faller categories, the Twoing splitting rule was used for CART analysis. This approach reduces bias toward the largest outcome strategy. Ten-fold cross validation was used to reduce the complicity of the prediction model (pruning). This procedure can prevent overfitted prediction models, thereby preserving generalizability.18 Logistic regression was used to evaluate the potential confounding and modification effects of treatment group assignment on the prediction model.
Four hundred eight participants, age 62.0±12.7 years, were assessed at 63.8±8.5 days poststroke and were monitored for falls incidence for 10.3±2.1 months. The average number of monthly reports per participant was 9.6±2.4; there was no significant difference in reporting across the 3 intervention groups (P=0.80). All participants had severe (<0.4 m/s; n=218; 53.4%) or moderate (0.4m/s≤ and <0.8 m/s; n=190; 46.6%) walking speed deficits, and most had moderate to moderately-severe disability (modified Rankin Scale [mRS]19 0–1, 0.5%; mRS 2, 13.2%; mRS 3, 42.2%; mRS 4, 44.1%).
Incidence of Falls
Among all participants, 147 (36.0%) were M/I, 88 (21.6%) S/NI, and 173 (42.4%) non-fallers. The majority of all fallers (n=235) experienced multiple falls (n=147; 62.6%). Twenty-four fallers (10.2%) experienced fall-related serious injury; 8 had one and 16 had >1 fall (P=0.43). For all falls (n=612), 55.4% occurred in the first 3 months monitored (3–5 months poststroke), 86.8% occurred at home (indoors, 70.8% [bathroom, 10.0%; bedroom, 23.9%; other room, 36.9%]; outdoors, 16.0%), and 13.1% in the community. Of individuals who fell, 74% had at least 1 fall from which they could not get up independently. Fall rate per person-year was 1.76 overall, 1.33 for moderately impaired walkers, and 2.13 for severely impaired walkers (P<0.001).
Characteristics of Fallers
The Table is an abbreviated summary of the associations between clinical presentation at 2 months and faller category at 12 months poststroke (online-only Supplemental Table; http://stroke.ahajournals.org). Multiple or injurious fallers were older than were S/NI fallers and nonfallers (P=0.02); S/NI and M/I fallers had worse upper (P=0.02) and lower (P=0.05) extremity motor control (Fugl-Meyer Upper and Lower Extremity Motor Scores,20 respectively) compared with nonfallers; and the M/I faller group had the lowest comfortable and fast walking speeds, 6-minute walk distance, and Berg Balance Scale (BBS) score; the greatest use of assistive devices; and the lowest balance confidence (P<0.01). The Stroke Impact Scale Participation subscale score was lowest (worst function) among S/NI and M/I fallers (P=0.01). Similarly, M/I fallers had the highest overall disability (lowest mRS scores), followed by S/NI fallers and nonfallers (P<0.01).
Impact of RCT Physical Therapy Interventions
There was no difference across intervention groups (HEP, early-LTP, and late-LTP) in overall fall incidence between 2 and 12 months poststroke. However, significantly more individuals who received early-LTP experienced M/I falls than did individuals who received late-LTP or HEP (P=0.047; Figure 1A). The difference in M/I falls across intervention groups is attributable to those with severe walking impairment (<0.4 m/s) in the early-LTP group (Figure 1B). Figure 2 provides log-rank test results and Kaplan-Meier estimates for probability of not having M/I falls from randomization to 12 months poststroke by intervention group and initial walking speed impairment.
A secondary analysis at 6 months poststroke demonstrated that early-LTP and HEP groups' M/I falls rates were 25.2% and 22.2%, respectively, compared with 14.0% for the late-LTP group, which received only usual care (P=0.05). However, as reported in a previous analysis,15 at 6 months, the late-LTP group was also less mobile (Stroke Impact Scale Mobility subscale score: early-LTP, 15.3±21.4; HEP, 14.9±20.0; late-LTP, 7.0±15.7) and took almost half as many steps in comparison with early-LTP and HEP (steps per day: early-LTP, 1017; HEP, 1357; late-LTP, 566).
Predicting Faller Category
Analysis using the CART method revealed that a BBS score ≤42 at 2 months poststroke was the single best predictor of M/I falls (Figure 3). Sensitivity and specificity were 73% and 53%, respectively, for the LEAPS cohort, and 78% and 39% with cross validation. Adding variables to the model enhanced prediction accuracy within the LEAPS cohort, but had poor generalizability with cross-validation. The association between BBS and fall category was neither confounded nor modified by intervention group assignment.
Incidence and Consequences of Falls
The LEAPS study provided a valuable opportunity to evaluate falls in a cohort of individuals receiving interventions to improve walking. The successful prospective capture of falls in this study reveals that even for individuals who are improving poststroke in mobility, balance, and walking,15 incidence of falls is high. Well over half of participants fell between 2 and 12 months poststroke. This high rate of falls is consistent with previous reports.4,6,7,21,22 Our data also support previous findings that individuals are most likely to fall at home,8,23,24 and that a large number experience falls from which they are unable to get up independently.4,23
Impact of RCT Physical Therapy Interventions
The LEAPS RCT assessed the impact of 2 physical therapy interventions on walking speed. Both LTP and HEP interventions were associated with clinically relevant improvements in walking speed, endurance, functional status, and quality of life in the severity groups categorized by either moderate or severe walking speed impairment.15
Between 2 and 6 months poststroke, both groups receiving early intervention had a higher fall rate than did individuals in the late-LTP group (who had received only usual care at 6 months)—this despite the fact that the early groups had experienced almost twice the improvement in walking and mobility as did the late-LTP group.15 The process of gaining mobility poststroke appears to be associated with higher risk for falls. Between 2 and 12 months poststroke, individuals with severe walking speed impairment in the early-LTP group experienced a significantly higher incidence of M/I falls than did those in the HEP group. We are uncertain about the causes of this difference. For those with severe walking speed impairment, locomotor training may have resulted in overconfidence in walking ability and/or perhaps HEP resulted in improved strength or balance.
Characteristics of Fallers and Predicting Faller Category
Our results corroborate others' findings that fall risk poststroke is associated with a wide range of characteristics including older age; greater disability; more prevalent use of an assistive device; and reduced balance, motor function, and walking speed.4,6,8,21–25 Unlike some reports,24,26 we did not find an association between falls and cognition or depression. This was probably attributable to the relatively high cognitive function (mean Mini Mental State Examination score, 26.1±3.5) and low incidence of depression (16.4% with Personal Health Questionnaire Depression Scale scores >9) in the cohort.15
The BBS emerged as the most robust predictor of M/I falls. Persson and colleagues22 identified the same BBS cutoff score for predicting fallers versus nonfallers in the first year poststroke. This convergence of findings suggests that the cutoff score of 42 may be a useful tool for identifying both single and M/I falls risk. Post test probability for M/I falls was 25% for BBS >42 and 42% for BBS ≤42. The low sensitivity and specificity for predicting M/I falls reflect the multifactorial causes of falls and suggest that a measure of balance, although useful, cannot independently account for fall risk.3 In addition, there may be items in the BBS that fail to distinguish between faller groups and detract from its predictive value. Identifying item subsets that are more robust for falls prediction may be useful.
The clinician and patients with stroke face a conundrum: walking may increase risk for M/I falls, whereas not walking will lead to a plethora of deficits associated with inactivity, including recurrent stroke.25 Clearly, there is a need for efficacious interventions that provide concomitant mobility training and fall prevention. For example, multirisk factor falls prevention programs should include Center for Disease Control-recommended home patient and caregiver education, a progressive exercise program, medication review and management, vision examination and improvement, and home safety assessment and modification.27 Attention to patient-specific deficits identified by the BBS may also be beneficial.
A strength of this study is that it was prospective and longitudinal with excellent falls capture rate in a well-defined cohort. The methods follow international standards for fall injury prevention trials.17 In addition, we were able to capture fall rates in association with interventions designed to improve mobility. The primary limitation of this study is the selective nature of the population. The population in this study was highly screened and living in the community; hence, results are not generalizable to a larger stroke population. For example, our cohort did not include nonambulatory individuals or individuals walking at speeds associated with community mobility (>0.8 m/s). Likewise, participants had no previous history of stroke or significant cardiac or neurological comorbidities and had relatively high cognitive function. Our analysis of risk factors was also limited to those collected for the RCT. Some risk factors (eg, history of falls and urinary incontinence) identified by others were not collected.
In a highly screened and selective population, individuals with stroke may improve in walking and mobility, but they remain at high risk for falls. Among those with severe impairment, participation in early-LTP appeared to increase M/I falls risk. The BBS is useful for informing fall risk, but has limitations related to the multifactorial nature of the problem. Fall prevention should focus on risk at home in the early months poststroke and should include multifactorial risk assessment and management concomitant with exercise interventions to improve walking and mobility.
Sources of Funding
This work was supported by funding from National Institute of Neurological Disorders and Stroke and the National Center for Medical Rehabilitation Research (RO1 NS050506).
LEAPS investigators include: Duke University Administrative Coordinating Center: Pamela Duncan, PT, PhD; Sarah Hayden; Mysha Sissine; Quishi Feng, PhD; Brooks Rehabilitation Hospital, Jacksonville, FL: Deborah Stewart, MD; Trevor Paris, MD; Joann Gllichio, PT, DSc; Florida Hospital, Orlando, FL: Mitchell Freed, MD; Michelle Dolske, PhD; Craig Moore, PT; Bettina Brutsch, PT; Long Beach Memorial Hospital, Long Beach, CA: H. Richard Adams, MD; Diehma Hoang, MD; Anita Correa, PT; Sharp Rehabilitation Center, San Diego, CA: Jerome Stenehjem, MD; Roxanne Hon, MD; Molly McLeod, PT; University of Southern California: David Alexander, MD; UCLA Medical Center: Julie Hershberg, DPT; Samneang Ith-Chang, DPT; Official Coordinating Center, University of Florida: Andrea L. Behrman, PT, PhD; Dorian K. Rose, PT, PhD; Clinical Coordinating Center, University of Southern California: Julie K. Tilson, DPT, MS; Data Management and Analysis Center, University of Southern California: Steven Cen, PhD; Chris Han, MS; James Gardner; University of Florida, Gainesville, FL: Yunfeng Dai, MS; Xiaomin Lu, PhD; Consultants: Anatole D. Martin, PhD, and Richard Schofield, MD, University of Florida; Steering Committee: Pamela Duncan, PT, PhD; Andrea L. Behrman, PT, PhD; Stanley P. Azen, PhD, and Samuel S. Wu, PhD, University of Southern California; Bruce H. Dobkin, MD, University of California, Los Angeles; Stephen Nadeau, MD, University of Florida; Sarah K. Hayden, Duke University.
The online-only Data Supplement is available at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.111.636258/-/DC1.
- Received September 12, 2011.
- Accepted October 26, 2011.
- © 2012 American Heart Association, Inc.
- Roger VL,
- Go AS,
- Lloyd-Jones DM,
- Adams RJ,
- Berry JD,
- Brown TM,
- et al
- Davenport RJ,
- Dennis MS,
- Wellwood I,
- Warlow CP
- Forster A,
- Young J
- Pouwels S,
- Lalmohamed A,
- Leufkens B,
- de Boer A,
- Cooper C,
- van Staa T,
- et al
- Ashburn A,
- Hyndman D,
- Pickering R,
- Yardley L,
- Harris S
- Nevitt MC,
- Cummings SR,
- Hudes ES
- Ada L,
- Dean CM,
- Morris ME,
- Simpson JM,
- Katrak P
- Bernhardt J,
- Dewey H,
- Thrift A,
- Collier J,
- Donnan G
- Boysen G,
- Krarup LH,
- Zeng X,
- Oskedra A,
- Kõrv J,
- Andersen G,
- et al
- Breiman L,
- Friedman JH,
- Olshen RA,
- Stone CJ
- Bonita R,
- Beaglehole R
- Kerse N,
- Parag V,
- Feigin VL,
- McNaughton H,
- Hackett ML,
- Bennett DA,
- et al
- Divani AA,
- Vazquez G,
- Barrett AM,
- Asadollahi M,
- Luft AR
- Jørgensen L,
- Engstad T,
- Jacobsen BK
National Center for Injury Prevention and Control. Preventing Falls: How to Develop Community-based Fall Prevention Programs for Older Adults. Atlanta, GA: Centers for Disease Control and Prevention, 2008.